The need for energy conservation is increasing the requirements for increased levels of electric motor efficiency. When operating at full load, induction motors have good efficiencies, even at relatively modest kW-ratings. However, at lower values than rated loads, which are a condition that many machines experience for significant portion of their service life, their efficiencies decrease somewhat. It is therefore important to optimize the efficiency of motor drive systems if significant energy savings are to be obtained. Several methods and techniques, that allow efficiency optimization when the induction motor works with light loads, have been reported, with different variables being controlled. Variable speed drive is one of the schemes that used to optimize the induction motor efficiency. As the load varies, the speed of the motor is varied. This way of controlling the motor speed with the load change has proven very suco applications and one of the most important applications in hot climate countries is in the field of Ventilation and Air-conditioning (VAC. ( The existing off-line method for improving IM motors efficiency are difficult to implement for on-line motors. The difficulty lies in the look-up tables, since they require more memory space and are expensive. This project presents a new application of Artificial Neural Network (ANN) control technique in the minimization of losses in an IM drive system. The ANN-based controller generates the appropriate voltage amplitude and frequency, which operate the IM at its optimum efficiency for any given operating point (torque and speed). The proposed ANN model is made of two networks connected in series: tansigmoidal and linear. A MATLAB program was developed for this purpose. The neural network model was trained using the MATLAB/NNET Toolbox based on the trainbpx.m program, which trains the network using the "backpropagation" method. The training data was generated by another MATLAB program which determines the voltage (rms) value and frequency (op, Vop) for given torque and speed (an, I'm) that operates the machine at optimum efficiency. After being trained appropriately, the ANN model reproduces the same output (fop; Vop) from a given input (Chs Tm), which lies between the minimum and maximum set points. The proposed method can be easily implemented on induction motor drive systems equipped with speed sensors and for which the steady state speed-vs-torque load characteristics are known. The suggested method will improve the efficiency of the drive system; hence it will save a considerable amount of energy. However, like many induction motor control techniques, the developed model is sensitive to motor parameter variations due to temperature change.